Hello Kafka Community,
We are investigating a recurring Kafka broker startup failure on Windows Server 
and would appreciate guidance from the Kafka maintainers and community.
Environment

  *
Apache Kafka: 4.0.1
  *
Operating system: Windows Server
  *
Java: Eclipse Temurin JDK 17.0.18
  *
Metadata mode: KRaft, single-node combined broker/controller
  *
Service management: NSSM
  *
Kafka storage directory: D:\KafkaStorage
  *
Configuration: D:\Kafka\config\kraft\server.properties

Problem description
Kafka repeatedly fails during startup while recovering KRaft metadata 
snapshots. The service manager reports the Windows service as Running or 
restarts the Java process, but Kafka does not successfully start its listeners. 
Ports 9092 and 9093 remain unavailable.
The recurring exception is:
java.nio.file.AccessDeniedException: 
D:\KafkaStorage\__cluster_metadata-0\00000000000004062214-0000001671.checkpoint.deleted
The stack trace indicates that the exception occurs during file deletion in:
kafka.raft.KafkaMetadataLog$.recoverSnapshots(KafkaMetadataLog.scala:691)
The failure is followed by a fatal process termination.
Investigations performed

  1.
Confirmed that Kafka is no longer running after the Windows service is stopped.
  2.
Checked the affected file's NTFS permissions. The SYSTEM account has Full 
Control.
  3.
Checked the file with Sysinternals Handle while Kafka was stopped; no matching 
open handles were found.
  4.
Verified that the D: volume is NTFS, reports Healthy/Operational OK, and is not 
marked dirty.
  5.
Found approximately 165 checkpoint files marked with the Windows ReadOnly 
attribute.
  6.
Clearing the ReadOnly attribute on one previously failing checkpoint file did 
not resolve startup; Kafka subsequently failed on another .checkpoint.deleted 
file.
  7.
Confirmed that the issue recurs and that restarting the Windows service alone 
does not resolve it.

Questions for the Kafka team

  1.
Is this a known issue in Kafka 4.0.1 on Windows when recovering KRaft metadata 
snapshots?
  2.
Does KafkaMetadataLog.recoverSnapshots() have any known Windows-specific file 
deletion or rename limitations?
  3.
Is the presence of numerous ReadOnly .checkpoint and .checkpoint.deleted files 
consistent with a Kafka defect, or should we investigate a specific Windows 
filesystem, security or file-attribute interaction?
  4.
Are there known fixes in later Kafka releases that address this failure path?
  5.
What additional diagnostics or a minimal reproduction would help determine 
whether the root cause is in Kafka, the JDK/Windows filesystem interaction, or 
our environment?

We want to identify the root cause and implement a permanent fix rather than 
rely on manual attribute changes or repeated service restarts.
We can provide the relevant configuration (with sensitive information 
redacted), complete startup logs, file attributes, ACL output, and additional 
diagnostics as needed.
Thank you for your guidance.
Regards,
Autonautics Engineering Team

Reply via email to